Resume interrupted runs

[SimonK04]

Idea: Resume interrupted runs (Checkpoint)

Problem

A run can take a long time and sends many requests to the API along the way.
If it stops in the middle (e.g. due to a crash, Ctrl+C, or an API error)
all the work done so far is lost. You have to start over from scratch, which
costs time and API budget.

Idea

AutoRecLab already saves the state of the tree search (as a pickled search
state). We could use this saved state to continue an interrupted run from
exactly the point where it stopped, instead of starting over.

For example via a flag:

autoreclab --resume <run-folder>

The agent then loads the last saved state and continues from there.

Benefits

• Saves time on long experiments
• Saves API costs, because already completed steps are not repeated
• Makes the tool more robust against crashes and connection errors

Questions for discussion

• Should resume happen automatically (the last run is simply continued), or
  explicitly via a flag (--resume)?
• How often should the state be saved: after every step, or only at certain
  intervals?
• Is it enough to continue from the last state, or would it be useful to also
  restart from an earlier point?